Date: Sat, 4 Mar 2023 10:57:48 -0600
From: Justin Pryzby <pryzby@telsasoft.com>
To: Jacob Champion <jchampion@timescale.com>
Cc: pgsql-hackers@postgresql.org, gkokolatos@pm.me,
	Michael Paquier <michael@paquier.xyz>,
	Tomas Vondra <tomas.vondra@enterprisedb.com>,
	Robert Haas <robertmhaas@gmail.com>, Peter Geoghegan <pg@bowt.ie>,
	Peter Eisentraut <peter.eisentraut@enterprisedb.com>,
	Heikki Linnakangas <hlinnaka@iki.fi>,
	Thomas Munro <thomas.munro@gmail.com>,
	Dipesh Pandit <dipesh.pandit@gmail.com>,
	Andrey Borodin <x4mmm@yandex-team.ru>,
	Mark Dilger <mark.dilger@enterprisedb.com>
Subject: Re: zstd compression for pg_dump
Message-ID: <20230304165747.GH12850@telsasoft.com>
References: <20230224191840.GD1653@telsasoft.com>
 <20230226012227.GK1653@telsasoft.com>
 <CAAWbhmgEpYPn2sjxr0Kar_XKS4dzyEnussd0emwe2YHxb_tk6g@mail.gmail.com>
 <20230303185546.GE12850@telsasoft.com>
 <CAAWbhmj-pEhpzfMA+ARnq+L3uS1A1AsmNjz9m_Cx=73n4GRZjQ@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAAWbhmj-pEhpzfMA+ARnq+L3uS1A1AsmNjz9m_Cx=73n4GRZjQ@mail.gmail.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Archived-At: <https://www.postgresql.org/message-id/20230304165747.GH12850%40telsasoft.com>
Precedence: bulk

On Fri, Mar 03, 2023 at 01:38:05PM -0800, Jacob Champion wrote:
> > > With this particular dataset, I don't see much improvement with
> > > zstd:long.
> >
> > Yeah.  I this could be because either 1) you already got very good
> > comprssion without looking at more data; and/or 2) the neighboring data
> > is already very similar, maybe equally or more similar, than the further
> > data, from which there's nothing to gain.
> 
> What kinds of improvements do you see with your setup? I'm wondering
> when we would suggest that people use it.

On customer data, I see small improvements - below 10%.

But on my first two tries, I made synthetic data sets where it's a lot:

$ ./src/bin/pg_dump/pg_dump -d pryzbyj -Fp -Z zstd:long |wc -c
286107
$ ./src/bin/pg_dump/pg_dump -d pryzbyj -Fp -Z zstd:long=0 |wc -c
1709695

That's just 6 identical tables like:
pryzbyj=# CREATE TABLE t1 AS SELECT generate_series(1,999999);

In this case, "custom" format doesn't see that benefit, because the
greatest similarity is across tables, which don't share compressor
state.  But I think the note that I wrote in the docs about that should
be removed - custom format could see a big benefit, as long as the table
is big enough, and there's more similarity/repetition at longer
distances.

Here's one where custom format *does* benefit, due to long-distance
repetition within a single table.  The data is contrived, but the schema
of ID => data is not.  What's notable isn't how compressible the data
is, but how much *more* compressible it is with long-distance matching.

pryzbyj=# CREATE TABLE t1 AS SELECT i,array_agg(j) FROM generate_series(1,444)i,generate_series(1,99999)j GROUP BY 1;
$ ./src/bin/pg_dump/pg_dump -d pryzbyj -Fc -Z zstd:long=1 |wc -c
82023
$ ./src/bin/pg_dump/pg_dump -d pryzbyj -Fc -Z zstd:long=0 |wc -c
1048267

-- 
Justin